CAEPR Working Paper #2006-012

Is Exporting a Source of Productivity Spillovers?

Roberto Alvarez

Central Bank of Chile

Ricardo Lopez Indiana University Bloomington

September 26, 2006

This paper can be downloaded without charge from the Social Science Research Network electronic library at: http://ssrn.com/abstract=932943.

The Center for Applied Economics and Policy Research resides in the Department of Economics at Indiana University Bloomington. CAEPR can be found on the Internet at: http://www.indiana.edu/~caepr. CAEPR can be reached via email at caepr@indiana.edu or via phone at 812-855-4050.

©2006 by Roberto Alvarez and Ricardo Lopez. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source.

Is Exporting a Source of Productivity Spillovers?*

Roberto Álvarez†

Ricardo A. López‡

September 2006

Abstract

This paper investigates whether exporting generates positive productivity spillover effects on

other plants operating in the same industry and whether exporting affects productivity of

plants in vertically related industries. Using plant-level data from Chile we find that exporters

improve productivity of their local suppliers but not of plants that purchase intermediate

inputs from them. We also find evidence of horizontal spillovers from exporting. Exporting by

foreign-owned plants generates positive spillovers in all directions: to their suppliers,

customers, and to other plants in the same industry. Domestic exporters increase productivity

of their suppliers and, to a lesser extent, that of plants in the same sector.

JEL: F10, F23, 03, 054

Keywords: exporting, spillovers, productivity, vertical linkages, Chile

* We would like to thank Hadi Salehi Esfahani, Gerhard Glomm, Holger Görg, Kim Huynh, Beata S. Javorcik, Renata Kosova, Volodymyr Lugovskyy, Mahmut Yasar and seminar participants at Illinois, Indiana, Memphis, the Midwest International Economics Meeting at Kansas, the International Industrial Organization conference at Northeastern University, and the European Trade Study Group Conference at the University of Vienna for very helpful comments and suggestions. We also thank Jean Morrison for proofreading and editing the manuscript. † Central Bank of Chile. E-mail: ralvarez@bcentral.cl. ‡ Corresponding Author. Department of Economics, Indiana University. Address: Wylie Hall Room 105, Indiana University, Bloomington, IN 47405, U.S.A. Telephone: +1-812-856-1466. Fax: +1-812-855-3736. E-mail: rialopez@indiana.edu.

1

1. Introduction

Many people believe that exporting firms generate knowledge about technologies and foreign

markets which can be used by other exporters and non-exporters in ways that increase their

productivity. Surprisingly, we know little about the effects of exporting on other firms’

productivity and whether foreign-owned exporters, domestic exporters, or both, are the ones who

generate spillovers. Using data for Chilean manufacturing plants, we investigate whether exporting

by both foreign-owned and domestic plants generates positive productivity spillover effects on

plants operating in the same industry and in vertically related industries.

Most of the previous literature studies the effect of general exporting activity on the

probability of exporting and export performance (e.g. Aitken, et al., 1997; Clerides, et al., 1998;

Barrios, et al., 2003; Bernard and Jensen, 2004a) but, with only few exceptions (e.g. Clerides, et

al., 1998; Javorcik, 2004; Girma et al., 2004; Görg and Hijzen, 2004), it has overlooked the effect of

exporting on productivity. Moreover, no study exists looking at productivity spillovers from

exporting by domestic plants.

In general, scholars find very little support for the idea that exporting increases the

probability of exporting and export performance of other firms. We believe that only looking at

the impact of spillovers on export performance may be misleading. Since there are sunk-entry costs

to export markets,1 it may be difficult to observe general exporting activity inducing entry to

export markets unless spillover effects are big enough to compensate for these entry costs.

Moreover, most of the studies look for intra-industry spillovers and ignore the potential linkages

from buyers of output to suppliers of inputs and vice versa.

1 See Roberts and Tybout (1997), and Bernard and Jensen (2004a).

2

From a policy point of view, it is important to analyze whether these spillover effects exist

or not. The existence of spillovers from exporting has been traditionally used as a justification for

the adoption of export promotion programs. Many countries in the world have encouraged exports

with the idea that they might fuel economic growth. Researchers have investigated whether these

export promotion programs are justified by testing the existence of learning-by-exporting.2 But

from a policy perspective, the relevant question is whether exporting generates spillovers to other

firms. The existence of learning by exporting itself is not necessarily a justification for export

promotion unless it can be shown that these learning effects spill over the rest of firms.

Our paper is related to the economic development literature which argues that export

activity may generate demonstration effects or provide new technologies that are not available for

domestic producers.3 This paper is also consistent with microeconomic evidence documenting that

exporters are more productive than non-exporters. Starting with the study by Bernard and Jensen

(1999) for the U.S., scholars have found evidence of productivity differentials in favor of

exporters.4 In the case of Chile, Álvarez and López (2005) show that after controlling for size and

foreign capital participation, exporters are 19 percent more productive in terms of total factor

productivity than non-exporters. These differentials make learning by domestic firms from highly-

productive exporters potentially important.

We make several contributions to the empirical literature. First, we test for the existence of

spillovers from exports on plant productivity. Second, we not only consider spillovers from plants

in the same industry, but also explore the role of vertically linked activities. Third, we analyze if

2 See recent surveys by López (2005), Greenaway and Kneller (2005), and Wagner (2005). 3 Some scholars, however, are more skeptical about the existence of these spillover effects. See Rodrik (1999), and Panagariya (2000). 4 See, for example, Bernard and Wagner (2001) for Germany; Isgut (2001) for Colombia; and Baldwin, and Gu (2003) for Canada. Wagner (2005) surveys the empirical strategies and results of 45 studies for 33 countries. He concludes that the evidence is robust in terms that exporters are more productive than non-exporters. Interestingly, most of these studies reveal that firms self-select in international markets while exporting does not necessarily have a positive effect on firm productivity (see also López, 2005).

3

there is a different impact between domestic and foreign-owned plants’ exports. By making this

distinction, we investigate if spillovers, as other authors have claimed, are mostly provided by

multinational enterprises. And fourth, we address several estimation issues that have plagued

previous studies. In particular, unlike previous works, we take into account the possible

endogeneity of our spillover variables by employing IV estimation methods. We construct three

different types of sector-level real exchange rates and use them as instruments. Our identification

assumption is that real exchange rate is correlated with industries export orientation, but it does

not affect plants productivity directly. In addition, following Aitken et al. (1997), we control for

general concentration of economic activity at region and industry level to make sure that we are

effectively capturing the impact of export activity, and not the impact of agglomeration or specific

advantages of some locations.

Using information for Chilean manufacturing plants from 1990 to 1999, we find strong

support for the view that exporters improve productivity of their local suppliers. We also find

evidence of horizontal spillovers from exporting but not from exporters to their customers.

Exporting by foreign-owned plants generates positive spillovers in all directions: to their suppliers,

customers, and to other plants in the same sector. Our finding that domestic exporters increase

productivity of their suppliers and, to a lesser extent, that of plants operating in the same industry

indicates that positive spillovers are not only associated with a larger presence of multinational

exporters, but also with exporting activity of domestic firms. Thus, we conclude that researchers

could have underestimated the role of domestic exporters in generating positive effects on other

firms’ productivity.

4

2. Spillovers from Exporting

The presumption that spillovers from exporting exist has been traditionally used as a

justification for the adoption of export promotion programs. Several arguments for why exporting

may generate these spillovers have been proposed. For example, consider a firm entering in a new

market or developing a new product for foreign markets; it faces several costs such as promotional

investments, making contacts with new clients, and technological innovation expenditures. Once

the firm achieves its objective, however, there is no impediment for other firms to enter this

market or imitate the new product without also paying these costs. This positive externality

suggests that investment in opening new markets and developing new products may be lower than

the socially optimal level (Westphal, 1990). Other authors argue that exporters tend to adopt

efficient and competitive management styles, and training of a higher quality of labor which may

benefit firms in other sectors (e.g. Keesing, 1967; Feder, 1982; Edwards, 1993).

The existence of these externalities and the role for export promotion, however, are highly

controversial. Advocates of active export promotion policies have used such justifications to

support government intervention. According to Lall (2002), the evidence suggests that export

promotion policies have been effective for improving export performance in newly industrialized

economies. Skeptics argue that these policies distort competition and undermine the multilateral

free trade system.5

Therefore empirical evidence on this regard is important to evaluate whether these

spillovers exist. Table 1 shows the studies that have studied the existence of spillovers from

exporting. Most of the studies explore potential technological or information spillovers from

5 Panagariya (2000), for example, discusses how traditional and recent arguments fail on theoretical and empirical grounds as justifications for the implementation of selective policies for export promotion, while Rodrik (1999) argues that there is not robust evidence of spillovers emanating from exporting activities.

5

exporters to other firms’ export performance. They analyze how export concentration affects the

probability of exporting and/or export intensity (measured as the export to sales ratio). These

analyses typically focus on firms operating in the same industry and/or region and in some cases

they distinguish between exports by domestic firms from exports by multinational corporations.

These studies either do not find evidence that export activity increases the probability of exporting

(e.g. Clerides et al., 1998; Barrios et al., 2003; Bernard and Jensen, 2004a) or find that only

multinational exporters generate spillovers (e.g. Aitken et al., 1997; Greenaway, et al., 2004;

Ruane and Sutherland, 2004). The effect of exporting activity on export intensity of exporters is

also not clear. While some find a positive effect of exporting activity by multinationals on export

intensity (e.g. Greenaway, et al., 2004) others find a negative effect (e.g. Ruane and Sutherland,

2004).

Table 1 also shows studies that have looked at productivity spillovers from exporting. Most

of them focus on foreign-owned exporters and consider the intra-industry aspect of spillovers. Only

Clerides et al. (1998) study the potential productivity spillovers from domestic exporting. But

their results do not provide support for their existence. Using Colombian plant-level data they find

that high export activity is not, in general, associated to lower production costs. In fact, in some

cases exporting appears to increase costs of production. As seen in the table, none of the studies

looks for spillover across sectors from domestic exporters through buyer-seller relationships. There

are several ways by which exporters may affect their suppliers (backward spillovers). They may

transfer knowledge and technically assist firms in upstream industries, so they can satisfy higher

quality requirements in foreign markets. In addition, an expansion of export industries may

increase the demand, or generate new demand, for intermediate inputs in upstream sectors.6

6 In Chile this seems to be the case with recent expansions in exports of wine and salmon. Once these industries maturated, there was a growing demand for specialized inputs.

6

There are also arguments favoring the existence of forward export spillovers. This would be

the case when downstream industries may become more productive as a result of gaining access to

new, improved, or less costly intermediate inputs. Although these spillovers have been commonly

associated to the presence of multinationals, there are no reasons to disregard that exporters may

be responsible for the same phenomenon. Consider, for example, the Chilean case of fruit exports.

Fruit is raw material for production of juice, canned fruit, and other more elaborate products. It is

reasonable that technological advances in industries producing the input or the introduction of a

new variety (raw fruit) may have an important effect on downstream industries (juice, canned

fruit).

The arguments presented in this section refer to positive spillovers. Theoretical

considerations, however, prevent us of being too optimistic. First, horizontal spillovers may be

unobserved in practice because firms have incentives to prevent information flows to competitors.

Second, export expansion in some regions or industries may increase the cost of labor or of other

specialized inputs. In these cases, the net spillover effect may be ambiguous. The net effect on

plant productivity then depends on the balance between the positive effect provided by

technological transfer and the negative effect of increased competition on input prices and the

scale of production.7

7 This negative effect has been denominated “congestion.” Evidence on this regard has been found by Karpaty and Kneller (2006) for the entry of multinationals in Sweden.

7

3. Data and Econometric Strategy

3.1 Data

The empirical analysis is based on the Annual National Industrial Survey (ENIA) carried out by

the National Institute of Statistics of Chile (INE) for the years 1990 through 1999. This survey

covers the universe of Chilean manufacturing plants with 10 or more workers. A plant is not

necessarily a firm; however, a significant percentage of firms in the survey are actually single-plant

firms (Pavcnik, 2002). The INE updates the survey annually by incorporating plants that started

operating during the year and excluding those plants that stopped operating for any reason.

For each plant and year, the ENIA collects data on production, value added, sales,

employment and wages (production and non-production), exports, investment, depreciation,

energy usage, foreign licenses, and other plant characteristics. In addition, plants are classified

according to the International Standard Industrial Classification (ISIC) rev 2. Using 4-digit

industry level price deflators, all monetary variables were converted to constant pesos of 1985.

Plants do not report information on capital stock, thus it was necessary to construct this variable

using the perpetual inventory method for each plant.

3.2 Econometric Strategy

We study the role of productivity spillovers from export activities by considering an augmented

production function which explicitly incorporates the role of spillovers:

(1) 10 1 2 3 2

3

ln( ) ln( )

+ ln( )

NP Pijrt ijrt ijrt ijrt jt jt

jt ijrt

y k l l Horizontal Backward

Forward

α α α α β ββ ε

= + + + + + +

+,

8

where ijrty is the log of value added of plant i operating in sector j and region r at time t; ijrtk is

the log of plant’s capital stock, while NPijrtl and P

ijrtl are the logs of non-production and production

labor respectively. The horizontal spillover variable for a given industry, say j, is defined as the

exports to sales ratio of that industry:

(2) ijt

i jjt

ijti j

ExportsHorizontal

Sales∈

∈

=∑∑

.

Thus, we are assuming that the larger the share of exports in a given industry, the larger

the potential spillover effect. The Backwardjt variable is a proxy for the export orientation of

industries that are supplied by industry j:

(3) ,

jt jk ktk k j

Backward Horizontalα≠

= ∑ ,

where jkα is the proportion of sector j’s output supplied to sector k. We calculate these

coefficients using data from the input-output matrix of Chile, constructed by the Central Bank of

Chile, at the 3-digit ISIC level for the year 1996. Given that we are interested in linkages within

the country and across productive sectors, we exclude the output for final consumption as well as

the imports of intermediate products. Finally, the Forwardjt variable attempts to measure the

export orientation of industries that supply inputs to industry j:

(4) ,

jt jk ktk k j

Forward Horizontalσ≠

= ∑ ,

where jkσ is the share of inputs purchased by industry j from industry k in total inputs purchased

by industry j.

Figure 1 shows the average value for the period 1990-1999 of the horizontal variable at the

3-digit sector level. As can be seen, the most export-oriented sectors are basic chemicals (351),

non-ferrous metals (372), paper (341), wood (331), and iron and steel (371), while sectors such as

9

non-metallic products (369), petroleum products (353, 354), plastic (356), and professional

equipment (385) export a very low fraction of their output.

Figures 2 and 3 show the backward and the forward variables, respectively. There are

important differences across industries. For example, the backward variable, which measures the

average export orientation of sectors that are supplied by the given industry, is high in ceramics

and glass (361, 362), plastic (356), and basic chemicals (351), but very close to zero for transport

equipment (384), footwear (324), and rubber products (355). The forward variable, which

measures the export orientation of sectors that provide inputs to the given industry, also varies

across sectors. High values are observed in printing (342), furniture (332), metal products (381),

leather products (323), and beverages (313), while low numbers are found in iron and steel (371),

non-ferrous metals (372), and wood products (331).

For estimation purposes, it will be convenient to re-write equation (1):

(5) 11 2 3 0 2

3

ln( ) ln( )

+ ln( )

NP Pijrt ijrt ijrt ijrt jt jt

jt ijrt

y k l l Horizontal Backward

Forward

α α α α β ββ ε

− − − = + + +

+.

The left-hand side of this equation is the traditional measure of the log of total factor

productivity (TFP) at the plant level. To measure TFP we estimate a Cobb-Douglas production

function for each 3-digit level industry using the method proposed by Olley and Pakes (1996) and

later modified by Levinsohn and Petrin (2003a, 2003b), which corrects the simultaneity bias

associated with the fact that productivity is not observed by the econometrician but it may be

observed by the firm (see Appendix for more details). The residuals of these regressions correspond

to our measures of productivity. Once TFP has been measured, we estimate the following

equation:

(6) 10 2 3ln( ) ln( ) ln( )ijrt jt jt jt ijrtTFP Horizontal Backward Forwardα β β β ε= + + + + .

10

There are several estimation issues that need discussion. First of all, there may be

unobserved plant characteristics which make some plants more productive. In that case the error

term in equation (6) can be decomposed into ijrt i ijrtc uε = + , where ic is the unobserved plant-

specific effect, and ijrtu is an error term. Then (6) becomes:

(7) 10 2 3ln( ) ln( ) ln( )ijrt jt jt jt i ijrtTFP Horizontal Backward Forward c uα β β β= + + + + + .

In the estimation, we treat ic as fixed effects and use OLS to estimate the parameters of

the within transformation of (7). Since there may be also sector, region, and year specific effects

that affect productivity we add a full set of 3-digit sector, region, and year dummy variables.

A second issue is that we need to control for the geographic concentration of the industry.

Suppose, for example, that plants tend to agglomerate in some sectors and regions.8 These

agglomeration effects may make plants that operate in that industry/region more productive and,

if the sector is also exporting a high fraction of their output, we may erroneously conclude that

exporting increases productivity of the plants. To control for this possibility, we include a measure

of the geographic concentration of the economic activity in the sector/region. We use two

measures of concentration:

Concentration 1

rjt

jtrjt

rt

t

EmploymentEmployment

EmploymentEmployment

⎛ ⎞⎜ ⎟⎝ ⎠=⎛ ⎞⎜ ⎟⎝ ⎠

,

and

Concentration 2

rjt

jtrjt

rt

t

Gross OuputGross Output

Gorss OutputGross Output

⎛ ⎞⎜ ⎟⎝ ⎠=⎛ ⎞⎜ ⎟⎝ ⎠

.

8 See Head and Mayer (2004) for a survey on agglomeration and trade.

11

A third estimation issue is a possible endogeneity of the spillover variables. Suppose, for

instance, that some sectors export more because the plants that operate in that sector are more

productive. Furthermore, some plants may increase their productivity with the purpose of

becoming exporters (Halward-Driemeier et al., 2002; López, 2005). Similarly, more productive

plants may self-select and supply inputs to sectors with a high export orientation. In these cases

the error term in equation (7), ijrtu , will be correlated with the spillover variables, so that the OLS

estimates will be inconsistent. To address this problem, we use the method of instrumental

variables. We instrument our three spillover variables using sector-level real exchange rates. We

assume that the level of the real exchange rates is correlated with the export shares but not with

variables other than exports that affect productivity (the error term in equation (7)). We argue

that this is a reasonable assumption for two reasons. First, there is plenty of evidence that

variations in real exchange rate are associated with significant changes in exports.9 Second, it

seems hard to argue that measures of real exchange rate at the industry-level can affect a variable

such as productivity that is measured at plant-level. Following recent models of firm heterogeneity

and international trade, we may expect a positive relationship between real exchange rate and

industry average productivity, but not for individual plants’ productivity. In fact, a real

depreciation may be thought of as a reduction in trade costs, which according to Melitz (2003) and

Bernard et al. (2006), raises the level of competition and the aggregate productivity of the

industry.

We construct three real exchange rates indices. The first one (RERjt) is a weighted average

of the bilateral real exchange rates between Chile and the 15 main destination countries of the

Chilean exports of the industry:

9 Recent evidence by Bernard and Jensen (2004b) show that real depreciations increase the export share of US plants in the manufacturing sector.

12

1

C

jt cj ctc

RER RERθ=

=∑ ,

where RERct is the bilateral real exchange rate between Chile and country c;10 C=15 is the number

of countries; and θcj is defined as:

1

1 Tcjt

cjt jt

ExportsT Exports

θ=

= ∑ ,

where Exportscjt is the value of exports from industry j to country c at time t; Exportsjt is the

value of exports from industry j at time t; and T is the number of periods trade data is available

(9 years, from 1991-1999). This index is assumed to be correlated with the export share of the

sector (the Horizontal variable).

The other two instruments measure the real exchange rate that exporters face in upstream

sectors (RER-Backwardjt) and the real exchange rate faced in downstream sectors (RER-

Forwardjt). They are defined following equations (3) and (4):

,jt jk kt

k k j

RER Backward RERα≠

− = ∑ , and

,jt jk kt

k k j

RER Forward RERσ≠

− = ∑ ,

where we are assuming that the higher the real exchange rate that exporters face in downstream

and upstream sectors, the higher the export share of those sectors.

We use these instruments to obtain predicted values of our three spillover variables, which

are then used to estimate the effect of exporting on plant productivity. The real exchange rates

turn out to be highly correlated with export shares at the sector level. A simple regression between

industries’ export share and real exchange rates, both in logs, gives us a coefficient of 1.34 and a t

10 The bilateral real exchange rate between Chile and country c is: RERct=NomERct*Pct/PChile,t. NomERct is the nominal exchange rate between Chile and country c (Chilean pesos / country’s c currency), while Pct and PChile,t are producer price level indices for country c and Chile, respectively. The nominal exchange rates and producer prices were obtained from the International Financial Statistics of the International Monetary Fund. In cases in which the producer price was not available the consumer price index was used.

13

statistic of 7.68. In order to check the validity of these instruments, we follow the traditional

procedures of looking at the individual t statistics for the coefficients of the three measures of

exchange rates, and the F statistics for the model including all the exogenous variables. The first-

stage regressions confirm that our instruments are adequate. The t statistics for the coefficient of

real exchange rates reveal that these variables are always significant at 1%. A more formal test is

the Anderson-Rubin test of the significance of the endogenous regressors.11 The null hypothesis

tested is that the coefficients of the endogenous regressors in the structural equation are jointly

equal to zero, and is numerically equivalent to estimating the reduced form of the equation (with

the full set of instruments as regressors) and testing that the coefficients of the excluded

instruments are jointly equal to zero. In all our estimations, the null hypothesis is rejected at 1%,

confirming the validity of our instruments.12 Figures 4, 5, and 6 are scatterplots of the true

spillover variables against their predicted values. These figures show that the real exchange rate

accounts for most of the variation in the export shares.

Table 2 shows descriptive statistics for all the relevant variables. There are 49,106 plant-

year observations, but after eliminating the ones for which we could not estimate TFP, we end up

with 40,476 observations.

4. Results

4.1 Basic Results

Table 3 reports our basic results of estimating equation (7). The first three columns of numbers

are the plant fixed effects estimates without taking into account the endogeneity problem. Column 11 This is different from Anderson-Rubin test for overidentifying restrictions. In our case, the model is exactly identified because we have three endogenous regressors and three excluded instruments. 12 The same conclusion is reached when we use as an alternative test of weak identification the Cragg-Donald test. All of these tests for first step regressions are generated by the command xtivreg2 in Stata.

14

(1) shows that the coefficient on backward and horizontal are positive, although only backward is

statistically significant. A 1% increase in the ratio exports/sales in downstream industries increases

productivity of plants in upstream industries in 0.291%, on average. Thus, sectors with higher

exports increase the productivity of plants that provide inputs to those sectors but do not increase

the productivity of plants that operate in the same industry. The forward variable is negative but

not significant. In columns (2) and (3) we control for the industry/region concentration of

economic activity. The labor concentration (concentration 1) is not significant and does not

change the estimates. The output concentration (concentration 2) is positive and statistically

significant suggesting that there may be some positive agglomeration externalities. The coefficients

for the spillover variables remain the same and the forward variable becomes marginally significant

at 10%.

In column (4) of Table 3 we present the estimates using the IV method with plant fixed

effects. All estimates are higher than the OLS estimates and now the horizontal variable is

statistically significant. A 1% increase in the exports/sales ratio increases productivity of plants in

the same industry by 0.05%, while productivity of plants in upstream industries increases by

0.52%. These results are robust to the inclusion of concentration measures (columns 5 and 6). In

sum, our evidence is consistent with the view that exporters provide positive spillovers to their

suppliers and to other plants in the same industry.

Why are the IV estimates higher than the OLS estimates? In our estimations, the export

share is used to proxy for the different ways in which interactions between plants raise

productivity (technical assistance to suppliers, demonstration effects, etc.). The export share is

likely to be correlated with these interactions but this correlation may be not perfect. Thus, in the

presence of measurement errors, OLS are biased downward. In addition, our export data at plant-

level comes from a survey, so the export data at industry-level may not coincide with the actual

15

amount exported. This suggests that the instrumented exports may be better predictors of the

spillover effects.

4.2 Who Generates Spillovers: Foreign-Owned or Domestic Exporters?

For a developing country, like Chile, it is possible that foreign-owned exporters are the main

source of technologies and knowledge. In other words, positive productivity spillovers may be more

likely to occur from exports by foreign-owned plants than from exports by domestic plants. To

analyze this possibility we split our spillover variables into two components: (1) exports by

foreign-owned plants; and (2) exports by domestic plants. Thus, we define the horizontal-foreign

spillover variable as:

ijt ijti j

jtijt

i j

F ExportsHorizontal Foreign

Sales∈

∈

− =∑∑

,

where Fijt is a dummy variable equal to one if plant i belonging to sector j has a positive amount

of foreign ownership at time t. In the same way we define the horizontal-domestic variable

considering exports by domestic plants only. The variables backward-foreign, backward-domestic,

forward-foreign and forward-domestic are defined following formulas (3) and (4).

Table 4 shows the results of estimating (7) using the exports of foreign-owned plants in our

spillover variables. Columns (1)-(3) refer to the case of OLS with plant fixed effects, while (4)-(6)

are the IV estimates with plant fixed effects. In all six cases the estimates for the three spillover

measures are positive and statistically significant, even when a concentration index is included. A

1% increase in the export/sales ratio increases productivity of plants in upstream industries by

0.16%-0.34%, in downstream sectors by 0.09%-0.27%, and in the same sector by 0.10%-0.23%.

16

These results give strong support to the idea that foreign-owned plants generate positive spillover

effects.

As a robustness check, we also estimate the effect of exporting by foreign-owned plants on

productivity of domestic plants only. The results, not presented here, are almost identical to those

in Table 4. This is consistent with the idea that affiliates of multinational corporations generate

positive productivity spillovers to domestic plants.

Do these findings mean that domestic exporters do not generate spillover effects? The

answer can be obtained from Table 5 which presents the estimates by using exports of domestic

plants only. We see that the estimate for backward is always positive and statistically significant.

A 1% increase in exports/sales increases productivity of plants in upstream sectors by 0.24%-

0.49%. For the forward and the horizontal variables the OLS and the IV regressions give slightly

different results. While the estimates for forward are negative in all six cases, they are not

significant when we use IV (columns 4-6). The horizontal variable, on the other hand, is positive

but never significant if we use OLS, and marginally significant at 10% when we use IV estimation.

There is then strong evidence that domestic exporters generate positive productivity spillovers to

their suppliers, some support for spillovers to other plants of the same industry but no evidence

that they benefit their customers.

In sum, our results suggest that positive spillovers are not only associated with a larger

presence of multinational exporters, but also with domestic exporters. In other words, by focusing

exclusively on foreign-owned firms, researchers have been underestimating the role of domestic

exporters in generating positive effects on other plants’ productivity.

17

5. Conclusions

Unlike most studies that have analyzed intra-industry or horizontal spillovers from export

activities, this paper focuses on inter-industry or vertical spillovers through backward (from

potential customers) and forward linkages (from potential suppliers). Anecdotal evidence suggests

that vertical spillovers, at least from exporters to their suppliers, may be important.

Using data from the manufacturing sector of Chile for the period 1990-1999, we confirm the

existence of positive productivity spillovers from exporters to their suppliers. This is evidence of

backward spillovers. We also find evidence that higher exporting activity in a given sector

increases the productivity of the plants operating in that sector. We do not find, however,

evidence of spillovers from exporters to their customers.

When we distinguish between foreign-owned plants exports and domestic plants exports we

discover that foreign-owned exporters generate positive productivity spillovers to their suppliers,

customers, and to other plants in the same industry. This is consistent with the perception that

multinational corporations transfer technologies in developing countries. But this does not mean

than domestic exporters do not improve the performance of other plants. We find strong support

for the existence of backward spillover effects from domestic exporters to their local suppliers and

some evidence that they benefit plants in the same sector.

Although we have been able to address several estimation issues that have plagued

previous studies such as the identification of spillover effects, the simultaneity problem, and the

role of unobserved plant characteristics, we still believe more work and better data are needed to

identify the exact mechanisms by which exporters transfer knowledge and technologies to other

firms operating either in the same industry or in other industries. Ideally, one would like to have

data on individual transactions between an exporter and its supplier and its customers.

18

Appendix: TFP Construction

To compute TFP we estimate a Cobb-Douglas production function separately for each industry.

Specifically, for each 3-digit sector, we estimate the following equation:

(A1) 0 1 2 3NP P

it it it it ity k l lβ β β β ε= + + + + ,

where ity is the log of value added of plant i at time t; itk is the log of plant's capital stock, while

NPitl and P

itl are the logs of non-production and production labor respectively. TFP is defined as:

( )1 2 3exp .NP Pit it it itTFP y k l lβ β β= − − −

If itε is uncorrelated with the right-hand side variables in equation (A1), then the

production function could be estimated using OLS. However, although productivity is not observed

by the econometrician it may be observed by the firm, thus itε is likely to be correlated with the

regressors. Following Olley and Pakes (1996), and Levinsohn and Petrin (2003a and 2003b) we

explicitly consider this endogeneity problem by writing it it itε ω η= + , where itω is the transmitted

productivity component and itη is an error term that is uncorrelated with input choices, and

assuming that ( , )it it it itm m k ω= , where itm is the intermediate input. Levinsohn and Petrin

(2003a) show that this relationship is monotonically increasing in itω , so the intermediate input

function can be inverted to obtain ( , )it it it itk mω ω= . Then, equation (A1) becomes:

(A2) 2 3 ( , )NP Pit it it it it ity l l k mβ β φ η= + + + ,

where 0 1( , ) ( , )it it it it it itk m k k mφ β β ω= + + .

Equation (A2) can be estimated using the procedures discussed in Petrin, Poi, and

Levinsohn (2004). As in Levinsohn and Petrin (2003a), we use consumption of electricity as the

intermediate input that allows the identification of the elasticity of capital.

19

REFERENCES

Aitken, B., Hanson, G. H., and A. E. Harrison (1997). “Spillovers, Foreign Investment and

Export Behavior,” Journal of International Economics 43(1-2): 103-132.

Álvarez, R., and R. A. López (2005). “Exporting and Performance: Evidence from Chilean

Plants,” Canadian Journal of Economics 38(4): 1384-1400.

Baldwin, J. R., and W. Gu (2003). “Export-Market Participation and Productivity

Performance in Canadian Manufacturing,” Canadian Journal of Economics 36(3): 634-

657.

Barrios, S., Görg, H., and E. Strobl (2003). “Explaining Firms’ Export Behavior: R&D,

Spillovers and the Destination Market,” Oxford Bulletin of Economics and Statistics

65(4): 475-496.

Bernard, A. B., and J. B. Jensen (1999). “Exceptional Exporter Performance: Cause, Effect,

or Both?” Journal of International Economics 47(1): 1-25.

Bernard, A. B., and J. B. Jensen (2004a). “Why Some Firms Export,” Review of Economics

and Statistics 86(2): 561-569.

Bernard, A. B., and J. B. Jensen (2004b). “Entry, Expansion, and Intensity in the US Export

Boom, 1987-1992,” Review of International Economics 12(4): 662-675.

Bernard, A. B., Redding, S. J., and P. K. Schott (2006). “Comparative Advantage and

Heterogeneous Firms,” Review of Economic Studies, forthcoming.

Bernard, A. B., and J. Wagner (2001). “Export Entry and Exit by German Firms,”

Weltwirtschaftliches Archiv 137(1): 105-23.

20

Clerides, S. K., Lach, S., and J. R. Tybout (1998). “Is Learning by Exporting Important?

Micro-Dynamic Evidence from Colombia, Mexico and Morocco,” Quarterly Journal of

Economics 113(3): 903-947.

Edwards, S. (1993). “Openness, Trade Liberalization, and Growth in Developing Countries,”

Journal of Economic Literature 31(3): 1358-1393.

Feder, G. (1982). “On Exports and Economic Growth,” Journal of Development Economics

12(1-2): 59-73.

Girma, S., Görg, H., and M. Pisu (2004). “The Role of Exporting and Linkages for

Productivity Spillovers from FDI,” GEP Research Paper 2004/30.

Görg, H., and A. Hijzen (2004). “Multinationals and Productivity Spillovers,” GEP Research

Paper 04/41, University of Nottingham.

Greenaway, D., and R. Kneller (2003). “Exporting, Productivity and Agglomeration: A

Difference in Difference Analysis of Matched Firms,” GEP Research Paper 03/45,

University of Nottingham.

Greenaway, D., and R. Kneller (2005). “Firm Heterogeneity, Exporting and Foreign Direct

Investment,” GEP Research Paper 05/32, University of Nottingham.

Greenaway, D., Sousa, N., and K. Wakelin (2004). “Do Domestic Firms Learn to Export

from Multinationals?” European Journal of Political Economy 20(4): 1027-1043.

Head, K. and T. Mayer (2004). “The Empirics of Agglomeration and Trade,” in V.

Henderson and J. F. Thisse (eds.), Handbook of Regional and Urban Economics, Vol. 4

(Amsterdam: North-Holland).

Isgut, A. (2001). “What's Different about Exporters? Evidence from Colombian

Manufacturing,” Journal of Development Studies 37(5): 57-82.

21

Javorcik, B. S. (2004). “Does Foreign Investment Increase the Productivity of Domestic

Firms? In Search of Spillovers through Backward Linkages,” American Economic Review

94(3): 605-627.

Karpaty, P., and R. Kneller (2005). “Demonstration or Congestion? Export Spillovers in

Sweden,” GEP Research Paper 05/44, University of Nottingham.

Keesing, D. B. (1967). “Outward-Looking Policies and Economic Development,” Economic

Journal 77(306): 303-320.

Kneller, R, and M. Pisu (2005). “Industrial Linkages and Export Spillovers from FDI,” GEP

Research Paper 05/20, University of Nottingham.

Lall, S. (2002). “Selective Policies for Export Promotion: Lessons from the Asian Tigers,” in

G. K. Helleiner (ed.), Non-Traditional Exports and Development in Sub-Saharan Africa:

Experience and Issues (London: Palgrave).

Levinsohn, J., and A. Petrin (2003a). “Estimating Production Functions Using Inputs to

Control for Unobservables,” Review of Economic Studies 70(2): 317-341.

Levinsohn, J., and A. Petrin (2003b). “On the Micro-Foundations of Productivity Growth.”

Mimeo, University of Chicago.

López, R. A. (2005). “Trade and Growth: Reconciling the Macroeconomic and Microeconomic

Evidence,” Journal of Economic Surveys 19(4): 623-648.

Olley, G. S., and A. Pakes (1996). “The Dynamics of Productivity in the

Telecommunications Equipment Industry,” Econometrica 64(6): 1263-1297.

Panagariya, A. (2000). “Evaluating the Case for Export Subsidies,” Policy Research Working

Paper 2276, World Bank, January.

Pavcnik, N. (2002). “Trade Liberalization, Exit, and Productivity Improvements: Evidence

from Chilean Plants,” Review of Economic Studies 69(1): 245-276.

22

Petrin, A., Poi, B. P., and Levinsohn, J. (2004). “Production Function Estimation in Stata

Using Inputs to Control for Unobservables,” Stata Journal 4(2): 113-123.

Roberts, M., and J. R. Tybout (1997). “The Decision to Export in Colombia: An Empirical

Model of Entry with Sunk Costs,” American Economic Review 87(4): 545-563.

Rodrik, D. (1999). New Global Economics and Developing Countries: Making Openness

Work. Policy Essay No. 24. Washington, D.C.: Overseas Development Council. Johns

Hopkins University Press, Baltimore.

Ruane, F., and Sutherland, J. (2005). “Foreign Direct Investment and Export Spillovers:

How Do Export Platforms Fare?” IIIS Discussion Paper No. 58, Institute for

International Integration Studies (IIIS), Trinity College Dublin.

Sjöholmm, F. (2003). “Which Indonesian Firms Export? The Importance of Foreign

Networks,” Papers in Regional Science, 82(3): 303-425

Wagner, J. (2005). “Exports and Productivity: A Survey of the Evidence from Firm Level

Data,” HWWA Discussion Paper No. 319.

Westphal, L. (1990). “Industrial Policy in an Export Propelled Economy: Lessons from South

Korea’s Experience,” Journal of Economic Perspectives 4(3): 41-59.

23

Table 1: Previous Studies on Exporting Spillovers Probability of

Exporting and/or Productivity

Export Intensity From Foreign-Owned Exporters

From Domestic Exporters

Horizontal AHH*, CLT, BGS,

GK, S, GSW, BJ, RS, KK, KP

CLT, GGP, GH CLT

Backward KP J, GGP None

Forward KP GGP None

* Study deals with endogeneity of industry/region export shares. AHH: Aitken, Hanson and Harrison (1997); CLT: Clerides, Lach and Tybout (1998); BGS: Barrios, Görg and Strobl (2003); GK: Greenaway and Kneller (2003); S: Sjöholmm (2003); GSW: Greenaway, Sousa and Wakelin (2004); BJ: Bernard and Jensen (2004a); RS: Ruane and Sutherland (2005); KK: Karpaty and Kneller (2005); KP: Kneller and Pisu (2005); GGP: Girma, Görg and Pisu (2005); GH: Görg and Hijzen (2004); J: Javorcik (2004).

24

Table 2: Descriptive Statistics

Number of

Observations Mean Std. Dev. Min Max

ln(TFP) 40,476 6.93 1.14 -4.57 12.74 ln(Horizontal) 49,106 -2.43 0.95 -5.70 -0.61 ln(Backward) 49,106 -4.78 1.29 -9.60 -2.90 ln(Forward) 49,106 -3.29 0.74 -5.62 -1.82 ln(Concentration 1) 49,106 0.12 0.68 -5.26 2.81 ln(Concentration 2) 49,106 0.13 0.94 -10.37 3.65 ln(RER) 49,106 4.58 0.15 3.65 4.76 ln(RER-Backward) 49,106 3.29 0.91 -0.37 4.29 ln(RER-Forward) 49,106 3.69 1.15 0.90 4.73 Concentration 1: Labor; Concentration 2: Gross Output.

25

Table 3: Productivity Spillovers from Exporting Plant Fixed Effects IV — Plant Fixed Effects (1) (2) (3) (4) (5) (6) Backward 0.291 0.291 0.291 0.519 0.519 0.527 (3.88)** (3.89)** (3.88)** (4.45)** (4.45)** (4.52)**Forward -0.092 -0.092 -0.094 -0.008 -0.008 -0.018 (1.68) (1.69) (1.72)+ (0.16) (0.16) (0.35) Horizontal 0.034 0.034 0.034 0.053 0.053 0.051 (0.93) (0.93) (0.92) (2.24)* (2.24)* (2.18)* Concentration 1 -0.003 0.002 (0.13) (0.16) Concentration 2 0.068 0.067 (4.20)** (6.27)**Number of Observations 40,476 40,476 40,476 39,648 39,648 39,648 R-Squared 0.183 0.183 0.184 0.177 0.177 0.178 Anderson-Rubin F-Stat 12.49** 12.51** 12.25** Cragg-Donald F-Stat 224.94** 224.94** 224.89** Notes: Absolute value of t statistics in parentheses (z statistics for IV regressions). Standard errors were clustered at the industry level in (1)-(3). Sector, region, and year dummy variables were included but not reported. + significant at 10%; * significant at 5%; ** significant at 1%. Concentration 1: Labor. Concentration 2: Gross Output. All variables in logs.

26

Table 4: Productivity Spillovers from Exporting by Foreign-Owned Plants

Plant Fixed Effects IV — Plant Fixed Effects (1) (2) (3) (4) (5) (6) Backward-Foreign 0.161 0.161 0.161 0.341 0.341 0.329 (3.26)** (3.26)** (3.25)** (4.45)** (4.44)** (4.31)**Forward-Foreign 0.091 0.091 0.091 0.262 0.261 0.266 (2.13)* (2.13)* (2.14)* (4.38)** (4.37)** (4.46)**Horizontal-Foreign 0.104 0.104 0.104 0.226 0.225 0.210 (2.67)* (2.67)* (2.70)* (2.68)** (2.68)** (2.50)* Concentration 1 0.000 0.010 (0.01) (0.65) Concentration 2 0.068 0.068 (4.14)** (6.37)**Number of Observations 40,476 40,476 40,476 39,648 39,648 39,648 R-Squared 0.184 0.184 0.185 0.169 0.169 0.171 Anderson-Rubin F-Stat 12.49** 12.51** 12.25** Cragg-Donald F-Stat 166.03** 166.79** 166.61** Notes: Absolute value of t statistics in parentheses (z statistics for IV regressions). Standard errors were clustered at the industry level in (1)-(3). Sector, region, and year dummy variables were included but not reported. + significant at 10%; * significant at 5%; ** significant at 1%. Concentration 1: Labor. Concentration 2: Gross Output. All variables in logs.

27

Table 5: Productivity Spillovers from Exporting by Domestic Plants

Plant Fixed Effects IV — Plant Fixed Effects (1) (2) (3) (4) (5) (6) Backward-Domestic 0.242 0.242 0.243 0.483 0.483 0.493 (2.97)** (2.98)** (2.98)** (3.78)** (3.77)** (3.86)**Forward-Domestic -0.105 -0.105 -0.107 -0.024 -0.024 -0.033 (2.37)* (2.37)* (2.42)* (0.47) (0.47) (0.65) Horizontal-Domestic 0.003 0.003 0.003 0.040 0.040 0.038 (0.11) (0.11) (0.11) (1.72)+ (1.71)+ (1.65)+Concentration 1 -0.001 0.004 (0.07) (0.26) Concentration 2 0.069 0.069 (4.22)** (6.40)**Number of Observations 40,476 40,476 40,476 39,648 39,648 39,648 R-Squared 0.182 0.182 0.183 0.174 0.174 0.175 Anderson-Rubin F-Stat 12.49** 12.51** 12.25** Cragg-Donald F-Stat 156.68** 155.82** 156.39** Notes: Absolute value of t statistics in parentheses (z statistics for IV regressions). Standard errors were clustered at the industry level in (1)-(3). Sector, region, and year dummy variables were included but not reported. + significant at 10%; * significant at 5%; ** significant at 1%. Concentration 1: Labor. Concentration 2: Gross Output. All variables in logs.

28

Figure 1: Horizontal Spillover Variable, 1990-1999

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

311 312 313 321 322 323 324 331 332 341 342 351 352 353 354 355 356 361 362 369 371 372 381 382 383 384 385 390

Figure 2: Backward Spillover Variable 1990-1999

0.000

0.005

0.010

0.015

0.020

0.025

0.030

0.035

0.040

0.045

0.050

311 312 313 321 322 323 324 331 332 341 342 351 352 353 354 355 356 361 362 369 371 372 381 382 383 384 385 390

Figure 3: Forward Spillover Variable 1990-1999

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

311 312 313 321 322 323 324 331 332 341 342 351 352 353 354 355 356 361 362 369 371 372 381 382 383 384 385 390

29

-6-4

-20

Hor

izon

tal

-5 -4 -3 -2 -1Predicted Horizontal

Figure 4: Actual vs. Predicted Horizontal Variable (In Logs)

-10

-8-6

-4-2

Bac

kwar

d

-10 -8 -6 -4 -2Predicted Backward

Figure 5: Actual vs. Predicted Backward Variable (In Logs)

30

-6-5

-4-3

-2F

orw

ard

-5 -4 -3 -2Predicted Forward

Figure 6: Actual vs. Predicted Forward Variable (In Logs)